PSNR Calculator — Peak Signal-to-Noise Ratio

PSNR (Peak Signal-to-Noise Ratio) measures how close a reconstructed signal — a compressed, denoised or super-resolved image — is to its original. It is the standard objective metric for image- and video-compression work, reported on a logarithmic decibel scale where a higher number means less distortion. This calculator follows the same definition used by scikit-image and MATLAB.

1. Find the peak value. The maximum representable pixel value is MAX = 2^bits − 1: 255 for an 8-bit image, 1023 for 10-bit, 4095 for 12-bit, 65535 for 16-bit.
2. Get the MSE. If you already know the mean squared error, enter it directly. Otherwise the tool computes it from your two pixel lists using MSE = (1/N)·Σ (xᵢ − yᵢ)² — the average of the squared differences between the original values xᵢ and compressed values yᵢ over N samples.
3. Apply the PSNR formula. PSNR = 10·log10(MAX² / MSE), measured in decibels. This is algebraically the same as 20·log10(MAX) − 10·log10(MSE), and the calculator shows that second form as an independent cross-check so you can see both agree to the rupee — here, to the decibel.
4. Handle the identical case. When the two signals match exactly, MSE = 0 and the ratio is divided by zero, so PSNR is mathematically infinite. Rather than erroring, the tool reports ∞ dB and labels it as lossless.

The coloured quality band (Excellent ≥ 40 dB, Good 30–40 dB, Acceptable 20–30 dB, Poor < 20 dB) is a widely-cited engineering heuristic for 8-bit images, useful for a quick read but not part of any formal standard — the page states this clearly next to the result. PSNR also has known limits: it scores pixel error, not perceived quality, so it can disagree with the eye. For perceptual work, pair it with SSIM or a learned metric. All maths here runs in your browser; nothing is uploaded.

• scikit-image — peak_signal_noise_ratio (reference implementation)
• MathWorks MATLAB — psnr function documentation
• Peak signal-to-noise ratio — definition and decibel form

The PSNR formula and the MAX = 2^bits − 1 convention on this page were last cross-checked against the scikit-image and MATLAB references on 2026-06-20. The quality bands are explicitly labelled as heuristics, not a standard.